A Lot Has Happened in A.I. Let’s Catch Up.

Thursday marked the one-year anniversary of the release of ChatGPT. A lot has happened since. OpenAI, the makers of ChatGPT, recently dominated headlines again after the nonprofit board of directors fired C. E. O. Sam Altman, only for him to return several days later. But that drama isn’t actually the most important thing going on in the A. I. world, which hasn’t slowed down over the past year, even as people are still discovering ChatGPT for the first time and reckoning with all of its implications. Tech journalists Kevin Roose and Casey Newton are hosts of the weekly podcast “Hard Fork.” Roose is my colleague at The Times, where he writes a tech column called “The Shift.” Newton is the founder and editor of Platformer, a newsletter about the intersection of technology and democracy. They’ve been closely tracking developments in the field since well before ChatGPT launched. I invited them on the show to catch up on the state of A. I. We discuss: who is — and isn’t — integrating ChatGPT into their daily lives, the ripe market for A. I. social companions, why so many companies are hesitant to dive in, progress in the field of A. I. “interpretability” research, and America’s “fecklessness” that cedes major A. I. benefits to the private sector, and much more. Recommendations: Electrifying America by David E. Nye Your Face Belongs to Us by Kashmir Hill “Intro to Large Language Models” by Andrej Karpathy (video) Import AI by Jack Clark. AI Snake Oil by Arvind Narayanan and Sayash Kapoor Pragmatic Engineer by Gergely Orosz Thoughts? Guest suggestions? Email us at [email protected]. You can find transcripts (posted midday) and more episodes of “The Ezra Klein Show” at nytimes.com/ezra-klein-podcast, and you can find Ezra on Twitter @ezraklein. Book recommendations from all our guests are listed at https://www.nytimes.com/article/ezra-klein-show-book-recs. This episode of “The Ezra Klein Show” was produced by Rollin Hu. Fact checking by Michelle Harris, with Kate Sinclair and Mary Marge Locker. Our senior engineer is Jeff Geld. Our senior editor is Claire Gordon. The show’s production team also includes Emefa Agawu and Kristin Lin. Original music by Isaac Jones. Audience strategy by Kristina Samulewski and Shannon Busta. The executive producer of New York Times Opinion Audio is Annie-Rose Strasser. And special thanks to Sonia Herrero.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:03Ezra Klein?

  From New York Times Opinion, this is the Ezra Klein Show.

• 0:22Ezra Klein?

  Before we get into the episode today, we are doing or getting ready to do our end of the year Ask Me Anything. So if you have questions you want to hear me answer on the show, I suspect a lot of them are going to be about Israel, Palestine and AI, but they don't have to be about Israel, Palestine and AI. Send them to ezra klein show at nytimes.com with AMA in the headline. Again, to ezra klein show at nytimes.com with AMA in the headline.

• 0:53Ezra Klein?

  If you follow business or tech or artificial intelligence news at all, in recent weeks, you certainly were following Sam Altman being unexpectedly fired a CEO of OpenAI, and then a huge staff revolt at OpenAI where more than 95% of the company said it would resign if he was not reinstated. And then he was reinstated. And so this whole thing seemed to have happened for nothing. I spent a lot of time reporting on this. And I talked to people on the Altman side of things. I talked to people on the board side of things. And the thing I am now convinced of, truly convinced of, is that there was less to it than met the eye. People saw, I saw, Altman fired by this nonprofit board meant to ensure

• 1:38Ezra Klein?

  that AI is built to serve humanity. And I assumed, and I think many assumed, there was some disagreement here over what OpenAI was doing, over how much safety was building into the systems, over the pace of commercialization, over the contracts it was signing, over what it was going to be building next year, over something. And that, I think I can say conclusively, and has been corroborated by other reporting, that was not what this was about. The OpenAI board did not trust and did not feel it could control Sam Altman. And that is why they fired Altman. It's not that they felt they couldn't trust him on one thing, that they were trying to control him on X, but he was beating them on X. It's that a lot of little things added up.

• 2:23Ezra Klein?

  They felt their job was to control the company, that they did not feel they could control him, and so to do their job, they had to get rid of him. They did not have, obviously, the support inside the company to do that. They were not ultimately willing to let OpenAI completely collapse. And so they largely, although I think in their view, not totally backed down. One of the members is still on the board. Altman and the president of OpenAI, Greg Brockman, are off the board. Some new board members are coming in who they think are going to be stronger and more willing to stand up to them. There's an investigation that is going to be done of Altman's behavior that will be at least released to the board, so they'll, I guess, know what to think of him. It's a very strange story. I wouldn't be surprised if there's things yet to come out, but I am pretty convinced

• 3:06Ezra Klein?

  that this was truly a struggle for control, not a struggle about X. But it has been a year since ChatGPT was released. It's a weird way to mark the year, but it has been a year. A year since OpenAI kicked off the whole modern era in artificial intelligence. A year since a lot of people's estimations of what humanity's future looked like began to shift and cloud and darken and shimmer. And so I wanted to have the conversation that many of us thought was a conversation happening here about what AI was becoming, how it was being used, how it was being commercialized, whether or not the path we're on is going to benefit humanity. And so I asked my friends over at Hard Fork, another great New York Times podcast, to come on the show. Kevin Roos is my colleague at The Times.

• 3:50Ezra Klein?

  He writes a tech column called The Shift. Casey Newton is the editor of Platformer, an absolutely must-read newsletter about the intersection of technology and democracy. And they have been following this in and out, but they've been closely following AI for the past year. So I wanted to have this broader conversation with them. As always, my email, [email protected].

• 4:18Ezra Klein?

  Kevin Rus, Casey Newton, welcome to the show, my friends.

• 4:21Speaker 01

  Hey, Ezra. Thanks for having us.

• 4:23Ezra Klein?

  All right, so we're talking on Monday, November 27th. ChatGPT, which kicked off this era in AI, was released on November 30th, 2022. So the big anniversary party was that Sam Altman got temporarily fired and the company almost collapsed and was rebuilt over at Microsoft, which I don't think is how people expected to mark the anniversary. But it has been now a year, roughly, in this sort of whole new AI world that we're in. And so I want to talk about what's changed in that year. And the place I want to begin is with the capabilities of the AI systems we're seeing, not the ones we're hearing about, but that we know are actually being used by someone in semi-real world conditions.

• 5:06Ezra Klein?

  What can AI systems do today that they couldn't do a year ago, Kevin?

• 5:12Speaker 01

  Well, the first most obvious capabilities improvement is that these models have become what's called multimodal. So a year ago, we had ChatGPT, which could take in text input and output other text as the response to your prompt. But now we have models that can take in text and output images, take in text and output video, take in voice data and output other voice data. So these models are now working with many more types of inputs and outputs than they were a year ago. And that's sort of the most obvious difference. If you just woke up from a year-long nap and took a look at the AI capabilities on the market, that's the thing that you would probably notice first.

• 5:58Ezra Klein?

  I want to pull something out about that. Is that it almost sounds like they're developing what you might call senses. And I recognize that there's a real danger of anthropomorphizing AI systems. So I'm not trying to do that. But one thing about having different senses is that we get some information that helps us learn about the world from our eyes, other information that helps us learn about the world from our ears, et cetera. One of the constraints on the models is how much training data they can have. As they become multimodal, it would seem that would radically expand the amount of training data. If you can have not just all of the text on the internet, but all of the audio on YouTube, or all the podcast audio on Spotify or something, or Apple Podcasts,

• 6:38Ezra Klein?

  that's a lot of data to learn about the world from, that in theory will make the models smarter and more capable. Does it have that kind of recursive quality?

• 6:47Speaker 01

  Absolutely. I mean, part of the backdrop for these capabilities improvements is this race for high-quality data. All of the AI labs are obsessed with finding new undiscovered high-quality data sources that they can use to train their models. And so if you run out of text because you've scraped the entire internet, then you've got to go to podcasts, or YouTube videos, or some other source of data to keep improving your models.

• 7:12Speaker 02

  For what it's worth, though, I don't think the availability of more training data is what is interesting about the past year. I think what was interesting about ChatGPT was that it gave average people a way to interact with AI for the first time. It was just a box that you could type in and ask it anything, and often get something pretty good in response. And even a year into folks using this now, I don't think we've fully discovered everything that it can be used for. And I think more people are experiencing Vertigo every day as they think about what this could mean for their own jobs and careers. So to me, the important thing was actually just the box that you type in and get questions from.

• 7:45Speaker 01

  Yeah, I agree with that. I think if you had just paused there, and there was no new development in AI, I think it would still probably take the next five or 10 years for society to adjust to the new capabilities in our midst.

• 7:59Ezra Klein?

  So you've made this point in other places, Casey, that a lot of the advances to come are going to be in user interfaces and in how we interact with these systems. In a way, that was a big advance of ChatGPT. The system behind it had been around for a while, but the ability to speak to it, or I guess write to it in natural language, it created this huge cultural moment around AI. But what can these AI products actually do that they couldn't do a year ago? Not just how we interface with them, but their underlying capacity or power?

• 8:30Speaker 02

  I mean, as of the Developer Day update that OpenAI had a few weeks back, the world knowledge of this system has been updated to April of this year. And so you're able to get something closer to real time knowledge of world events. It has now integrated with Microsoft Bing, and so you can get truly real time information in a way that was impossible when ChatGPT launched. And these might sound like relatively minor things, Azura, but you start chaining them together. You start building the right interfaces and you actually start to see beyond the internet as we know it today. You see a world where the web, where Google is not our starting point for doing everything online. It is just a little box on your computer that you type in and you get the answer without

• 9:15Speaker 02

  ever visiting a webpage. So that's all going to take many years to unfold, but the beginnings of it are easy to see now.

• 9:22Speaker 01

  One other capability that didn't exist a year ago, at least in any public products, is the ability to bring your own data into these models. So Claude was the first language model that I used that had the ability to, say, upload a PDF. So you could say, here's a research paper. It's 100 pages long. Help me summarize and analyze this. And it could do that. Now ChatGPT can do the same thing. And I know a bunch of other systems are moving in that direction, too. There are also companies that have tried to spin up their own language models that are trained on their own internal data. So if you are, you know, Coca-Cola or BCG or some other business and you want an internal ChatGPT that you can use for your own employees to ask, say, questions about your HR documents,

• 10:06Speaker 01

  that is a thing that companies have been building. So that's not the sexiest, most consumer facing application, but that is something that there's enormous demand for out there.

• 10:14Ezra Klein?

  So one thing it seems to me to be getting better at, from what I can tell from others, is coding. I have to ask people whether they're using AI bots very often and if so, for what? And basically, nobody says yes unless they are a coder. Everybody says, oh, yeah, I played around with it. I thought it was really cool. I sometimes use Dolly or Mid Journey to make pictures for my kids or for my, you know, email newsletter. But it is the coders who say, I'm using it all the time. It has become completely essential to me. I'm curious to hear a bit about that capability increase.

• 10:47Speaker 01

  I think where it has sort of become part of the daily habit of programmers is through tools like GitHub Copilot, which is a basically chat GPT for coders that finishes whatever line of code you're working on, or helps you debug some code that's broken. And there have been some studies and tests. I think there was one test that GitHub itself ran, where they gave two groups of coders the same task. And one group was allowed to use GitHub Copilot and one group wasn't. And the group with GitHub Copilot finished the task 55% faster than the group without it. Now, that is like a radical productivity increase. And if you tell a programmer, here's a tool that can make you 55% faster, they're going to want to use that every day.

• 11:32Ezra Klein?

  So when I see functionally chatbots in the wild, what I see is different versions of what people used to somewhat derisively call like the fancy autocomplete, right? Help you finish a line of code, help you finish this email. You ask a question that you might ask the search engine, like, why do I have spots all over my elbow? And it gives you an answer that hopefully is right, but maybe is not right. I do think some of the search implications are interesting. But at the same time, it is not the case that Bing has made great strides on Google. People have not moved to asking the the kind of Bing chatbot its questions as opposed to asking Google. Everybody feels like they need AI in their thing now, right? There's a, I don't think you can raise money in Silicon Valley at the moment if you don't have a generative AI play built into your product or built

• 12:17Ezra Klein?

  into your business strategy. But that was true for a minute for crypto, too. And I'm not one of the people who makes a crypto AI analogy. I think crypto is largely vaporware and AI is largely real. But Silicon Valley is faddish and people don't know how to use things. And so everybody tries to put things in all at once. What product has actually gotten way better?

• 12:38Speaker 02

  I'll just use one example. There's an app you might be familiar with called Notion. It's productivity collaborative software. I write a newsletter. I save every link that I put in my newsletter into Notion. And now that there is AI inside Notion, Notion can do a couple of things. One, it can just look at every link I save and just write a two sentence summary for me, which is just sort of nice to see at a glance what that story is about. And most recently, it added a feature where you can just do Q and A with a database and say like, hey, like, what are like some of the the big stories about Meta over the past few weeks? And it'll just start pulling those up, essentially querying the database that I have built. And so while we're very early in this, you're beginning to see a world where AI is taking data that you have stored somewhere, and it's turning it into your

• 13:23Speaker 02

  personal research assistant. So is it great right now? No, I would give it like a C. But for one point, no, I think it's not bad.

• 13:31Speaker 01

  And I'll share another example that is not from my own use, but I was talking a few weeks ago with a doctor. He's a friend of a friend. And doctors, you get tons of messages from patients, you know, what's this rash? Can you renew this prescription? Do I need to come in for a blood test? Like that kind of stuff. And doctors and nurses spend a ton of time just opening up their message portal, replying to all these messages. It's a huge part of being a doctor, and it's a part that they don't like. And so this doctor was telling me that they have this software now that essentially uses a language model, I assume it's OpenAI's or someone very similar to that, that goes in and prefills the responses to patient queries. And the doctor still has to look it over, make sure everything's right, and press send. But just that act of pre-populating the field, this

• 14:16Speaker 01

  person was saying it saves them a ton of time, like on the order of several hours a day. And if you have that, and you sort of extrapolate to, what if every doctor in America was saving themselves an hour or two a day of responding to patient messages? I mean, that's a radical productivity enhancement. And so you can say that that's just fancy autocomplete. And I guess on some level it is. But just having fancy autocomplete in these paperwork-heavy professions could be very important.

• 14:44Ezra Klein?

  Well, let me push that in two directions, because one direction is that I am not super thrilled about the idea that my doctor theoretically here is glancing over things and clicking submit as opposed to reading my message themselves and having to do the act of writing, which helps you think about things and thinking about what I actually emailed them and like what kind of answer they need to give me. I mean, I know personally the difference in thought between scanning things and editing and thinking through things. So that's like my diminishing response. But the flip of it is the thing I'm not hearing anybody say here, and the thing I keep waiting for and being interested in, is the things AI might be able to do better than my doctor. I was reading Jack

• 15:27Ezra Klein?

  Clark's Import AI newsletter today, which I super recommend to people who want to follow advancements in the field. And he was talking about a, I mean, it was a system being tested, not a system that is in deployment, but it was better at picking up pancreatic cancer from certain kinds of information than than doctors are. And I keep waiting to hear something like this going out into the field, right? Something that doesn't just save people a bit of time around the edges. I agree, that's a productivity improvement. It's fine. You can build a business around that. But the promise of AI, when Sam Altman sat with you all a few weeks ago, or however long it was, and said, we're moving to the best world ever, he didn't mean that our paperwork is going to get a little bit easier to complete. Like, he meant we'd have cures for new diseases. He meant that we would have new kinds of

• 16:08Ezra Klein?

  energy possibilities. I'm interested in the programs and the models that can create things that don't exist.

• 16:17Speaker 02

  Well, to get there, you need systems that can reason. And right now, the systems that we have just aren't very good at reasoning. I think that over the past year, we have seen them move a little away from the way that I was thinking of them a year ago, which was a sort of fancy autocomplete, right? It's sort of making it, making a prediction about what the next word will be. It's still true that they do it that way, but it is able to create a kind of facsimile of thought that can be interesting in some ways. But you just can't get to where you're going, Ezra, with like a facsimile of thought. You need something that has improved reasoning capabilities. So maybe that comes with the next generation frontier models, but until then, I think you'll be disappointed.

• 16:54Ezra Klein?

  But do you need a different kind of model? This is something that lingers in the back of my head. So I did an interview on the show with Demis Hassabis, who's the co-founder of DeepMind, now runs the integrated DeepMind Google AI program. And DeepMind had built this system a while back called AlphaFold, which treated how proteins are constructed in 3D space, which is to say, in reality, we live in 3D space. It treated it as a game, and it fed itself a bunch of information, and it became very good at predicting the structure of proteins, and that solved this really big scientific problem. And they then created a subsidiary of Alphabet called Isomorphic Labs to try to build drug discovery on similar foundations. But my understanding is that Google, during this period, became terrified of Microsoft and OpenAI beating

• 17:38Ezra Klein?

  it up in search and office. And so they pulled a lot of resources, not least Hassabis himself, into this integrated structure to try to win the chatbot wars, which is now what their system Bard is trying to do. And so when you said, Casey, that we need things that can reason, I mean, maybe, but also you could say, we need things that are tailored to solve problems we care about more. And I think this is one of the things that worries me a bit, that we've backed ourselves into business models that are not that important for humanity. Is there some chance of that? I mean, are we going too hard after language-based general intelligent AI that, by the way, integrates very nicely into a suite of enterprise

• 18:19Ezra Klein?

  software, as opposed to building things that actually create scientific breakthroughs, but don't have the same kind of high scalability profit structure behind them?

• 18:30Speaker 01

  I would stick up for the people who are working on the sort of what you could call like the non-language problems in AI right now. This stuff is going on. It maybe doesn't get as much attention from people like the three of us as it should. But if you talk to folks in fields like pharmaceuticals and biotech, there are new AI biotech companies spinning up every day, getting funding to go after drug discovery or some more narrow application. Like, we talked to a researcher the other day, formerly of Google, who is teaching AI to smell, right? Taking the same techniques that go into these transformer-based neural networks like ChatGPT, and applying them to the molecular structures of different chemicals, and using that to be able to predict what these

• 19:15Speaker 01

  Things will smell like. And you might say, well, what's the big deal with that? And the answer is that some diseases have smells associated with them that we can't pick up on because our noses aren't as sensitive as, say, dogs or other animals. But if you could train an AI to be able to recognize scent molecules and predict odors from just chemical structures, that could actually be useful in all kinds of ways. So I think this this kind of thing is happening. It's just not sort of dominating the coverage the way that ChatGPT is.

• 20:07Ezra Klein?

  Let me ask you, Kevin, about, I think, an interesting, maybe promising, maybe scary avenue for AI that you possibly personally foreclosed, which is, at some point during the year, Microsoft gave you access to a OpenAI-powered chatbot that had this dual personality of Sydney. And Sydney tried to convince you you didn't love your wife and that you want to run away with Sydney. And my understanding is immediately after that happened, everybody with enough money to have a real business model and AI lobotomized the personalities of their AI. It's like that was the end of Sydney. But there are a lot of startups out there trying to do AI friends, AI therapists, AI sex bots, AI boyfriends and girlfriends and non-binary partners,

• 20:51Ezra Klein?

  just every kind of AI companion you can imagine. I've always thought this is a pretty obvious way this will affect society. And the Sydney thing convinced me that the technology for it already exists. So where is that?

• 21:09Ezra Klein?

  And how are those companies doing?

• 21:11Speaker 01

  Yeah, I mean, I'm sorry, A, if I did foreclose the possibility of AI personalities. I think what's happening is it's just a little too controversial and sort of fraught for any of the big companies to wade into. Like, Microsoft doesn't want its AI assistants and co-pilots to have strong personalities. Like, that much is clear. And I don't think their enterprise customers want them to have strong personalities, especially if those personalities are adversarial or confrontational or creepy or unpredictable in some way. They want like, they want clippy, but like with real brain power. But there are companies that are going after this more social AI market.

• 21:52Speaker 01

  One of them is this company, Character AI, which was started by one of the original people at Google who made the transformer breakthrough. And that company is growing pretty rapidly. They've got a lot of users, especially young users. And they are doing essentially AI personas. You can make your own AI persona and chat with it, or you can pick from ones that others have created. Meta is also going a little bit in this direction. They have these sort of persona-driven AI chatbots. And all of these companies have put sort of guardrails around, like, no one really wants to do the erotic, what they call erotic sort of role play, in part because they don't want to run afoul of things like the Apple App Store Terms of Service.

• 22:36Speaker 01

  But I expect that that will also be a big market for young people. And anecdotally, I mean, I have just heard from a lot of young people who already say, like, my friends have AI chatbot friends that they talk to all the time. And it does seem to be making inroads into high schools. And that's just an area that I'll be fascinated to track.

• 22:56Speaker 02

  I mean, this is gonna be huge. A couple of thoughts coming to mind. One, I talked to somebody who works at one of the leading AI companies, and they told me that 99% of people whose accounts they remove, they remove for trying to get it to write text-based erotica. Okay? So that, I think, speaks to the market demand for this sort of thing. I've also talked to people who have used the models of this that are not constrained by any sort of safety guidelines, and I've been told these things are actually incredible. Writing erotica. So what I'm telling you is there is a 10 billion

• 23:22Ezra Klein?

  You've done really a lot of reporting on this case.

• 23:24Speaker 02

  I'm so interested in this.

• 23:25Ezra Klein?

  You say maybe a personal interest?

• 23:26Speaker 02

  Look, I write about content moderation and like porn is the content moderation frontier. And it's just very interesting to me that it's so clear that there are billions of dollars to be made here. And no company will touch it. And I asked one person involved. I said, why don't you just let people do this? And they basically said, look, if you do this, you become a porn company overnight. It like overwhelms the usage. Like this is what people wind up using your thing for. And you're you're working at a different company then. So I sort of get it. But you know, even setting aside the explicitly erotic stuff, you know, Ezra, you well know, and have talked and written about just like the loneliness epidemic that we have in this country. There's a lot of isolated people in this world. And I think there is a very real possibility that a lot of those people will find comfort and joy and delight with talking to these AI based companions. I also think that when that happens, there will be a culture war over it.

• 24:11Speaker 02

  And we will see lengthy segments on Fox News about how the Silicon Valley technologists created a generation of shut ins who wants to do nothing but talk to their fake friends on their phones. So I do think this is like the culture war yet to come. And the question is just sort of when do the enabling technologies get good enough? And when do companies decide that they're willing to deal with the blowback?

• 24:28Ezra Klein?

  I also think this is going to be a generational thing. I mean, I'm very interested in this and have been for for a bit. In part because I suspect if I had to make a prediction here, my five year old is going to grow up with AI friends. And my sort of pat line is that today we worry that 12 year olds don't see their friends enough in person. And tomorrow we'll worry that not enough of our 12 year olds friends or persons. Because it's going to become normal. And my sense is that the systems are really good. If you unleashed them, you're already good enough to functionally master this particular application. And the big players simply haven't unleashed them. I've heard from people at the big companies here who are like, oh yeah, if we wanted to do this, we could dominate it. But that does bring me to a question, which is Meta kind of does want to do this.

• 25:13Ezra Klein?

  Meta, which owns Facebook, which is a social media company. They seem to want to do it in terms of these lame seeming celebrity avatars. Like you can talk to AI Snoop Dogg.

• 25:23Speaker 02

  They're so bad.

• 25:24Ezra Klein?

  But Meta is interesting to me because their their AI division is run by Yann Le Cun, who's one of the most important AI researchers in the field. And they seem to have very different cultural dynamics in their AI shop than Google Deep Mind or OpenAI.

• 25:41Ezra Klein?

  Tell me a bit about Meta's strategy here and what makes them culturally different.

• 25:46Speaker 01

  Well, Casey, you cover Meta and have for a long time and may have some insight here. My sense is that they are sort of up against a couple of problems. One of which is they have arrived to AI late and degenerative AI specifically. You know, Facebook was for many years considered one of the top two labs along with Google when it came to recruiting AI talent, to putting out cutting edge research, to presenting papers at the big AI conferences. They were one of the big dogs. And then they sort of had this funny thing happen where they released a model called Galactica just right before ChatGPT was released last year. And it was supposed to be this sort of like LLM for science and for research papers.

• 26:28Speaker 01

  And it was out for, I think, three days and people started noticing that it was making up fake citations. It was hallucinating. It was doing what all the AI models do, but it was it was from Meta and so it felt different. It had sort of this tarnish on it because people already worried about fake news on Facebook. And so it got pulled down and then ChatGPT, just shortly thereafter, launched and became this global sensation. So they're sort of grappling for what to do with this technology that they've built now. There's not a real obvious business case for shoving AI chatbots into products like Facebook and Instagram. And they don't sell enterprise software like Microsoft does, so they can't really shove it into paid subscription products.

• 27:12Speaker 01

  So my sense from talking with folks over there is that they're just kind of not sure what to do with this technology that they've built. And so they're just flinging it open to the masses. What do you think?

• 27:22Speaker 02

  That tracks with me. I sort of basically don't get it either. Like, basically what you've just said has been explained to me. They are investing a ton with no obvious return on investment in the near term future. I will say that these celebrity AI chatbots they've made are quite bad. Like, it's truly baffling. And the thing is, they've taken celebrities, but the celebrities are not playing themselves in the AI. They've given all of the celebrities silly names and you can just sort of like follow their Instagram and like set Then messages and say like, hey, like, character that Snoop Dogg is portraying like, what do you think about it? So it's all very silly and I expected it'll die a rapid death sometime in the next year and then we'll see if they have a better idea. What I will say is like, if you're somebody who wakes up from AI nightmares some mornings, as a lot of folks in San Francisco

• 28:07Speaker 02

  do, go listen to Jan LeCoq and talk about it. No one has ever been more relaxed about AI than Jan LeCoq. You know, it's just sort of like an army of superhuman assistants are about to live inside your computer. They're gonna, you know, do anything you want to do and there's no risk of them harming you ever. So if you're, you know, you're feeling anxious, go, go listen to Jan.

• 28:24Ezra Klein?

  Do you think he's right? Because it also has led to policy difference. Meta has been much more open source in their approach, which OpenAI and Google seem to think is irresponsible. But there's something happening there that I think is also built around a different view of safety. Like, what is their view of safety? Why does Jan LeCoq, who is like an important figure in this whole world, why is he so much more chill than, you know, name your other founder?

• 28:49Speaker 02

  I mean, part of it is, I just think these are deeply held convictions from someone who is an expert on this space and who has been a pioneer and who understands the technology certainly far better than I do. And he can just sort of not see from here to killer robots. So I, I respect his viewpoint in that respect, given his credentials in the space. I think on the question of is open source AI safer, this is still an open question, not to pun. The argument for it being safer is, well, if it's open source, that means that average people can go in and look at the code and identify flaws and kind of see how the machine works. And they can point those out in public and then they can be fixed in public. Whereas if you have something like OpenAI, which is building very powerful systems behind closed doors, we don't have the same kind of

• 29:31Speaker 02

  access. And so you might not need to rely on a government regulator to see how safe their systems were. So that is the argument in favor of open source. Of course, the flip side of that is like, well, if you take a very powerful open source model and you put it out on the open web, even if it's true that anyone can poke holes and identify flaws, it's also true that a bad actor could take that model and then use it to do something really, really bad. So that hasn't happened yet, but it certainly seems like it's a obvious possibility at sometime in the near future.

• 29:58Ezra Klein?

  Let me use that as a bridge to safety more generally. So we've talked a bit about where these systems have gone over the past year, where they seem to be going. But there's been a lot of concern that they are unsafe and fundamentally that they become misaligned, or that we don't understand them or what they're doing. What kind of breakthroughs have there been with all this investment and all this attention on safety, Kevin?

• 30:20Speaker 01

  So a lot of work has gone into what is called fine tuning of these models. So basically, if you're making a large language model like GPT-4, you have several phases of that. Phase one is what's called pre-training, which is sort of just the the basic process. You take all of this data, you shove it into this neural network, and it learns to make predictions about the next word in a sequence. Then from there, you do what's called fine tuning. And that is basically where you are trying to turn the model into something that's actually useful or tailored, turn it into a chatbot, turn it into a tool for doctors, turn it into something for social AIs. That's the process that includes things like reinforcement learning

• 31:01Speaker 01

  from human feedback, which is how a lot of the leading models are fine tuned. And that work has continued to progress. The models, they say, today are are sort of safer and less likely to generate harmful outputs than previous generations of models. There's also this field of interpretability, which is where I've been doing a lot of reporting over the past few months, which is this sort of tiny subfield of AI that is trying to figure out what the guts of a language model look like and what what is actually happening inside one of these models when you ask it a question or give it some prompt and it produces an output. And this is a huge deal, not only because I think people want to know how these things work. They're not satisfied by just saying these are like mystical black boxes,

• 31:46Speaker 01

  but also because if you understand what's going on inside a model, then you can understand if, for example, the model starts lying to you or starts becoming deceptive, which is a thing that AI safety researchers worry about. So that process of of interpretability research, I think, is really important. There have been a few sort of minor breakthroughs in that field over the past year, but it is still slow going and it's still a very hard problem to crack.

• 32:10Speaker 02

  And I think it's worth just pausing to underscore what Kevin said, which is the people building these systems do not know how they work. They know at a high level, but there is a lot within that where if you show them an individual output from the AI, they will not be able to tell you exactly why it said what it said. Also, if you run the same query multiple times, you'll get slightly different answers. Why is that? Again, the researchers can't tell you. So as we have these endless debates over AI safety, one reason why I do tend to lean on the side of the folks who are scared is this exact point. At the end of the day, we still don't know how the systems work.

• 32:44Ezra Klein?

  Tell me if this tracks for you. I think compared to a year ago, when I talked to the AI safety people, the people who worry about AIs that become misaligned and do terrible civilizational level damage, AIs that could be really badly misused, they seem to think it has been actually a pretty good year, most of them. They think they've been able to keep big models like GPT-4, which of course are much less powerful than what they one day expect to invent, but they think they've been pretty good at keeping them aligned. They have made some progress on interpretability, which wasn't totally clear, you know, a year ago, many people said that was potentially not a problem we could solve, you know, at least we're making some breakthroughs there. They're not relaxed, the people who worry

• 33:27Ezra Klein?

  about this. And they, you know, will often say like, we would need a long time to fully understand even the things we have now, and we may not have that long. But nevertheless, I get the sense that the safety people seem a little more confident that the work, the technical work they've been doing is paying off than, you know, at least was the impression I got from the reporting prior.

• 33:50Speaker 02

  I think that's right. I mean, Sam Altman, in particular, this has been his strategy, is like, we are, we are going to release this stuff that is in our labs, and we're going to kind of wait and see how society reacts to it. And then we'll sort of give it some time to let society adjust, and then we will release the next thing. That's what he thinks is the best way to slowly integrate AI into our lives. And if you'd asked me maybe 11 months ago, like a month into using ChatGPT, what are the odds of something really, really bad happening because of the availability of ChatGPT? I would have put them much higher than they turned out to be, right? And when you talk to folks at OpenAI, like, they will tell you that that company really has taken AI safety really seriously.

• 34:32Speaker 02

  You can see this yourself when you use the product, ask it a question about sex, it basically calls the police. So there is a lot to be said for how these systems have been built so far.

• 34:42Speaker 01

  And I would say that the other thing that I've heard from AI safety researchers is that they are feeling relief, not just that the world has not ended, but that more people are now worried about AI. It was a very lonely thing for many years to be someone who worried about AI safety, because there was no apparent reason to be worried about AI safety, right? That the chatbots that were out, it was like Siri and Alexa, and they were terrible. And no one could imagine that these things could become dangerous or harmful, because the technology itself was just not that advanced. Now, you have congressional hearings, you have regulations coming from multiple countries. You have people like Jeff Hinton and Yashua Bengio,

• 35:26Speaker 01

  two of the so-called godfathers of deep learning, proclaiming that they are worried about where this technology is headed. So I think for the people who've been working on this stuff for a long time, there is some just palpable relief at like, oh, I don't have to carry this all on my shoulders anymore. The world is now aware of these systems and what risks they could pose.

• 35:44Ezra Klein?

  One irony of it is that my read from talking to people is that AI safety is going better as a technical matter than was expected. And I think worse as a matter of governance and inter-corporate competition and regulatory arbitrage than they had hoped. There's a fear, as I understand it, that we could make the technical breakthroughs needed, but that the kind of coordination necessary to go slow enough to make them, like, that's where a lot of the fear is. I think they feel like that's actually going worse, not better.

• 36:19Speaker 02

  So, one of the big narratives coming out of Sam Altman's firing was that it must have had something to do with AI safety. And, you know, based on my reporting and reporting shared by many others, this was not an AI safety issue. But it is very much the story that is being discussed about the whole affair. And the folks who are on the board, who are associated with these AI safety ideas, they've taken a huge hit to their public reputation because of the way they handled the firing and sort of all of that. And so I think a really bad outcome of this firing is that the AI safety community loses its credibility, even though AI safety, as far as we can tell, really didn't have a lot to do with what happened to Sam Altman.

• 37:01Ezra Klein?

  Yeah, I first agree that clearly AI safety was not behind whatever disagreements Altman and the board had. I heard that from both sides of this, and I didn't believe it, and I didn't believe it, and I finally was convinced of it. I was like, you guys had to have had some disagreement here. It seems so fundamental. But this is sort of what I mean, the governance is going worse. All the OpenAI people thought it was very important, and Sam Altman himself talked about its importance all the time, that they had this nonprofit board connected to this non-financial mission, right? The values of building AI that served humanity that could fire Sam Altman at any time, or even shut down the company fundamentally, if they thought it was going awry in some way or another. And the moment that board tried to do that, now, I think they did not try to do that on very strong grounds, but

• 37:46Ezra Klein?

  the moment they tried to do that, it turned out they couldn't. That the company could fundamentally reconstitute itself at Microsoft, or that, you know, the board itself couldn't withstand the pressure coming back. I think the argument from the board's side, the now mostly defunct board, is that this didn't go as badly for them as the press is reporting, that they brought in some other board members who are not cronies of Sam Altman and Greg Brockman. Sam Altman and Greg Brockman are not on the board now. There's going to be investigation into Altman. So maybe they have a stronger board that is better able to stand up to Altman. That is one argument I have heard. On the other hand, those stronger

• 38:28Ezra Klein?

  board members do not hold the views on AI safety that the board members who left, like Helen Tonner of Georgetown and Tasha McCauley from RAND, held. I mean, these are people who are going to be very interested in whether or not OpenAI is making money. I'm not saying they don't care about other things, too, but these are people who know how to run companies. They serve on corporate boards in a normal way, where like the output of the corporate board is supposed to be shareholder value, and that's going to influence them, even if they understand themselves to have a different mission here. I mean, am I getting that story wrong to you?

• 39:00Speaker 01

  No, I think that's right. And it speaks to one of the most interesting and sort of strangest things about this whole Industry is that the people who started these companies were weird. And I, and I say that with no sort of like normative judgment, but they made very weird decisions. Like, they thought AI was exciting and amazing, that they wanted to build AGI, but they were also terrified of it, to the point that they developed these elaborate safeguards. I mean, in OpenAI's case, they put this nonprofit board in charge of the for-profit subsidiary and gave essentially the nonprofit board the power to push a button and shut down the whole thing if they wanted to. At Anthropic, one of these other AI companies, they are structured as a public benefit corporation,

• 39:45Speaker 01

  and they have kind of this, their own version of a nonprofit board that is capable of essentially pushing the big red shut it all down button if things get too crazy. This is not how Silicon Valley typically structures itself. Like, Mark Zuckerberg was not in his Harvard dorm room building Facebook, thinking like, if this thing becomes the most powerful communication platform in the history of technology, like, I will need to put in place these checks and balances to keep myself from becoming too powerful. But that was the kind of thing that the people who started OpenAI and Anthropic were thinking about. And so I think what we're seeing is that that kind of structure is sort of bowing to the requirements of shareholder capitalism, which says that, you know, if you do need

• 40:29Speaker 01

  all this money to run these companies, to train these models, you are going to have to make some concessions to the sort of powers of the shareholder and of the money. And so I think that one of the big pieces of fallout from this OpenAI drama is just that OpenAI is going to be structured and run much more like a traditional tech company than this kind of holdover from this nonprofit board.

• 40:49Speaker 02

  And that is just a sad story. I truly wish that it had not worked out that way. I think one of the reasons why these companies were built in this way was because it just helped them attract better talent. I think that so many of people working in AI are idealistic and civic-minded and do not want to create harmful things, and they're also really optimistic about the power that good technology has. And so when those people say that as powerful and good as these things could be, it could also be really dangerous, I take them really seriously, and I want them to be empowered. I want them to be on company boards. And those folks have just lost so much ground over the past couple of weeks. And it is a truly tragic development, I think, in the development of this industry.

• 41:52Ezra Klein?

  One thing you could just say with that is, yeah, it was always going to be up to governments here, not up to strange nonprofit, corporate, semi corporate structures. And so we actually have seen a huge amount of government activity in in recent weeks. And so I want to start here in the US. Biden announced a big package of a big executive order. You could call them regulations. I would sort of call them pre regulations. But Casey, how would you describe in some what they did? Like, what is the Biden administration's approach that it is signaling to regulating AI?

• 42:27Speaker 02

  The big headline was if you are going to train a new model, so a sort of successor to a GPT four, and it uses a certain amount of energy and the energy there is just sort of a proxy for how powerful and capable this model might be. You have to tell the federal government that you have done this and you have to inform them what safety testing you did on this model before releasing it to the public. So that is the one kind of break that they attempted to put on the development of this industry. It does not say you can't train these models. It doesn't specify what safety tests you have to do. It just says if you're going to go down this road, you have to be in touch with us.

• 43:08Speaker 02

  And that will, I think, slightly decelerate the development of these models. I think critics would say it also pushes us a little bit away from a more open source version of AI, that open source development is sort of chaotic by its nature. And if you want to do some sort of giant open source project that would compete with the GPTs of the world, that would just sort of be harder to do. But to me, those are sort of the the big takeaways.

• 43:35Ezra Klein?

  One of the things that struck me looking at the order was, go back a year, go back two years, I think the thing that people had said is that the government doesn't understand this at all. It can barely be conversant in technology. People remember Senator Orrin Hatch asking Mark Zuckerberg, "Well, if you're not making people pay, then how do you make money?" When I read the order and looked at it, this actually struck me as pretty seriously engaged. Like, for instance, there's a big debate in the AI world about whether or not you're going to regulate based on the complexity and power of the model or the use of the model. You know, you have a fear about what happens if you're using the model for medical decisions, but if you're just using it as your personal assistant,

• 44:19Ezra Klein?

  who cares? Whereas like the AI safety people have the view that, no, no, no, the personal assistant model might actually be the really dangerous one because that's the one that knows how to act in the real world. And the Biden administration takes a real, like, it actually takes a view of the AI safety people, right? If you have a model over a certain level of computing complexity, they want this higher level of scrutiny, higher level of disclosure on it. They want everything that comes from an AI to be watermarked in some way so you can see that it is AI-generated. This struck me as a Biden administration actually clearly having taken this seriously and having convened some set of group of stakeholders and experts that knew what they were doing. I mean, I don't necessarily agree with literally every decision, and a lot of it is just asking for reports. But when you think about it as a framework for regulation, it didn't read to me as a framework

• 45:04Ezra Klein?

  coming from people who had not thought about this for 10 minutes.

• 45:07Speaker 02

  Absolutely. I was quite impressed. You know, I had a chance to meet with Ben Buchanan at the White House who worked on this, talked to him about this stuff, and it is clear that they have been reading everything, they've been talking to as many people as they can, and they did arrive in a really nuanced place. And I think when you look at the reaction from the AI developers in general, it was mostly like neutral to lightly positive, right? There was not a lot of blowback. But at the same time, folks in civic society, I think, were also excited that the government did have a point of view here and had done its own work.

• 45:36Speaker 01

  Yeah, it struck me as a very deft set of, I think I would agree that they're more like pre-regulations than regulations. And to me, it sounded like what the Biden White House was trying to do was throw a few bones to everyone. What was like, we're going to throw a few bones to the AI safety community who worries about foundation models becoming too powerful. We're going to throw some bones to the AI harms community that is worried about things like bias and inaccuracy. And we're going to throw some bones to the people who worry about foreign use of AI. So I saw it as a very sort of deliberate attempt to give every sort of camp in this debate a little to feel happy about.

• 46:15Ezra Klein?

  One of the things it raised for me as a question, though, was, did it point to a world where you think that regulators are going to be empowered to actually act? This was the thing I was thinking about after the board collapse. You imagine a world sometime in the future where you have OpenAI, you know, with GPT-6 or Meta or whomever, right? And they are releasing something that the regulators, looking at the safety data, looking at what's there, they're just itchy about it. It's not obviously going to do a ton of harm, but they're not convinced it's safe. They've seen some things that worry them. Are they really going to have the power to say, no, we don't think your safety testing

• 47:00Ezra Klein?

  was good enough, when this is a powerful company, when they won't be able to release a lot of the proprietary data, right? The thing where the board could not really explain why they were firing Sam Altman struck me as almost going to be the situation of virtually every regulator, trying to think about the future harms of a model. If you're regulating in time to stop a thing from doing harm, it's going to be a judgment call. And if it's a judgment call, it's going to be a very hard one to make. And so if we ever got to the point where somebody needed to flip the switch and say, no,

• 47:29Speaker 01

  Does anybody actually have the credibility to do it? Or is what we've seen that in fact, like these very lauded, successful companies run by smart people who have huge Twitter followings or threads followings, whatever they end up being on, that they actually have so much public power that they'll always be able to make the case for themselves? And like, the political economy of this is actually that we better just hope the AI companies get it right because nobody's really going to have the capability to stand in front of them.

• 47:58Speaker 02

  When you talk to folks who are really worried about AI safety, they think that there is a high possibility that at some point in, let's say, the next five years, AI triggers some sort of event that kills multiple thousands of people. What that event could be, you know, we could speculate. But assume that that is true. I think that changes the political debate a lot, right? Like, that's just all of a sudden, you start to see jets get scrambled. Hopefully, that never happens, but I think that will be the inciting moment. And this is the thing that just frustrates me as somebody who writes about tech policy, is we just live in a country that doesn't pass laws. There are endless hearings, endless debates, and then it gets time to regulate something. And it's like, well, yeah, they can regulate AI, but it's going to be based on this one, you know, regulation that was passed to deal with, like, you know, the oat farming crisis of 1906, and we're just going to hope that it applies. It's like, we should pass new laws in this country.

• 48:41Speaker 02

  I don't know that there's a law that needs to be passed today to ensure, you know, that all of this goes well, but certainly Congress is going to need to do something at some point as this stuff evolves.

• 48:49Speaker 01

  I mean, one thing I was thinking about as this whole situation at OpenAI was playing out was actually the financial crisis in 2008 and the scenes that were captured in books and movies where, you know, you have the heads of all the investment banks and they're scrambling to avoid, you know, going under and they're meeting in these boardrooms with people like Ben Bernanke, the Chair of the Federal Reserve. And the government actually had a critical role there in patching together the financial system because they were sort of interested, not in which banks survived and which failed, but in making sure that there was a banking system when the markets opened the next Monday. And so I think we just need a new regulatory framework that does have some kind of, the

• 49:34Speaker 01

  sort of cliched word would be stakeholder, but someone who is in there as a representative of the government who is saying, what is the resolution to this conflict that makes sense for most Americans or most people around the world?

• 49:47Ezra Klein?

  When you looked at who the government gave power to in this document, when you think about who might play a role like that, when you need to call the government on AI, the way I read it is it spread power out across a lot of different agencies. And there were places where invested more rather than less. But one thing that different people have called for that I didn't see it do, in part because you would actually need to pass a law to do this, was actually create the AI department, something that is funded and structured and built to do this exact thing, to be the central clearinghouse inside the government, to be led by somebody who would be the most credible on these issues and would maybe have then the size and strength to do this kind of research, right?

• 50:30Ezra Klein?

  The thing that is in my head here, because I find your analogy really compelling, Kevin, is the Federal Reserve. The Federal Reserve is a big institution, and it has a significant power of its own in terms of setting interest rates. It also does a huge amount of research. Like, when you think about where would a public option for AI come from? You would need something like that, that has the money to be doing its own research and hiring, you know, really excellent people, in that case economists, in this case AI researchers. And there was nothing like that here. It was sort of an assertion that we more or less have the structure we need, we more or less have the laws we need, we can apply all those things creatively. But it did not say, like, this is such a big deal that we need a new institution to, you know, be our point person on it.

• 51:15Speaker 01

  Yeah, I mean, I think that's correct. I think there are some reasons for that. But I think, you know, you do want a government that has its own technological capacity when it comes to AI. Previous waves of innovation, you know, certainly nuclear power during the Manhattan Project, but also things like the internet came out of DARPA. These are areas where the government did have significant technical expertise and was building its own technology in sort of competition with the private sector. There is no public sector equivalent of ChatGPT. The government has not built anything even remotely close to that. And I think it's worth asking why that is and what would need to happen for the government to have its own capacity, not just to evaluate and regulate these systems,

• 52:00Speaker 01

  but to actually build some of their own.

• 52:03Ezra Klein?

  I think it is genuinely strange on some level that, given how important this is, there is not a bill gathering steam. Look, if the private sector thinks it is worth pumping 50 or 100 billion dollars into these companies so they can help you make better enterprise software, it seems weird to imagine that there are not public problems that have an economic value that is equal to that or significantly larger. And we may just not want to pay that money, fine, but you know, we do that for infrastructure, right? I mean, we just passed a gigantic infrastructure bill. And if we thought of AI like infrastructure, we actually also spend a lot of money on

• 52:48Ezra Klein?

  broadband now. It seems to me you want to think about it that way. And I think it is a kind of fecklessness and cowardice on the part of like the political culture that it no longer thinks itself capable of doing things like that. Like, at the very least, and I've said this, I think, on your show probably, I think they should have prize systems where they say a bunch of things they want to see solved. And if you can build an AI system that will solve them, they'll give you a billion dollars. But the one thing is the government does not like to do things that spend money for an uncertain return. And building a giant AI system is spending a lot of money for an uncertain return. And so the only part of the government that is probably doing something like it is the Defense Department in areas that we don't know. And that does not make me feel better, that makes me feel worse.

• 53:33Ezra Klein?

  That's my take on that.

• 53:34Speaker 01

  Yeah, I mean, I think there's also a piece of this that has to do with labor and talent. You know, there are probably on the order of several thousand people in the world who can oversee the building, training, fine tuning, deployment of large language models. It is a very specific skill set and the people who have it can make gobs of money in the private sector, working wherever they want to. The numbers that you hear coming out of places like OpenAI for what engineers are being paid there, I mean, it's like NFL level football compensation packages for some

• 54:12Speaker 01

  Of their people and the government simply can't or won't pay that much money to someone to do equivalent work for the public sector. Now, I'm not saying they should be paying engineers millions of dollars of taxpayer money to to build these things, but that's that's what you would need to do if you wanted to compete in an open market for the top AI talent.

• 54:31Ezra Klein?

  I am saying they should. I am saying this is like the factlessness and cowardice point. This is stupid.

• 54:36Speaker 01

  You you think there should be AI engineers working for the federal government making five million dollars a year?

• 54:41Ezra Klein?

  Look, maybe not five million dollars a year, but it would. This is a this thing that we don't think civil servants should make as much as people in the private sector. Because I don't know, somebody at a congressional hearing is going to stand and be like, that person's making a lot of money. That is a way we rob the public of value. If Google's not wrong, Microsoft is not wrong, that you can create things that are of social value through AI. And if you believe that, then leaving it to them, I mean, they intend to make a profit. Why shouldn't the public get great gains from this? It won't necessarily be through profit.

• 55:19Ezra Klein?

  But, you know, if we could cure different diseases or, you know, make big advances on energy, I just this way of thinking is actually, to me, the the really significant problem. I'm not sure you would need to pay people as much as you're saying, because I actually do think a lot of, I mean, we both know the culture of the AI people, and at least up until a year or so ago. It was weird and a lot of them would do weird things and are not living very lush lives. Um, you know, they're in group houses with each other, taking psychedelics and working on AI on the on the weekdays. But I think you could get people in to do important work and you should. Now, look, you don't have the votes to do stuff like this. I think that's the real answer. But in other countries, they will and do like when Saudi Arabia decides it needs an AI.

• 56:02Ezra Klein?

  To be geo strategically competitive, it will take the money it makes from selling oil to the world. And in the same way that it's currently using that money to hire sports stars, it will hire AI engineers for a bazillion dollars and it will get some of them. And then it will have a decent AI system one day. I don't know why we're waiting on other people to do that. We're rich. It's stupid.

• 56:23Speaker 02

  I agree with you, Ezra, and I'm sorry that Kevin is so resistant to your ideas because I think paying public servants well would do a lot of good for this country.

• 56:31Speaker 01

  Look, I think public service should be paid well. I'm just saying, when Jim Jordan gets up and grills the former, you know, DeepMind engineer about why the Labor Department is paying them $2.6 million to fine tune language models, I'm not sure what the answer is going to be.

• 56:47Ezra Klein?

  No, I agree with you. I mean, but this, I think we're all saying in a way the same thing. It's like, this is a problem. Government by dumb things Jim Jordan says is not going to be a great government that takes advantage of opportunities for the public. Good. And that sucks. It would be better if we were doing this differently. And if we thought about it differently. Let me ask about China. Because China is where, on the one hand, at least on paper, the regulations look much tougher. So one version is maybe the regulating AI much more strictly than we are. Another view that I've heard is that, in fact, you know, that's true for companies, but the Chinese government is making sure that it's building very, very strong AI. You know, to the extent you all have looked at it, how do you understand the Chinese regulatory approach and how it differs from our own?

• 57:33Speaker 02

  I mean, I've looked at it mostly from the standpoint of what are the consumer facing systems look like. It has only been, I think, a couple of months since China approved the first consumer usable ChatGPT equivalent. As you might imagine, they have very strict requirements as far as like what the chatbot can say about Tiananmen Square. So they wind up being more limited maybe than what you can use in the United States. As far as what is happening behind closed doors and for their, their defense systems and that sort of thing, I'm in the dark.

• 58:08Speaker 01

  So four, five years ago, when I started reporting a book about AI, the conventional wisdom among AI researchers was that China was ahead and they were going to make all of the big breakthroughs and beat the US technology companies when it came to AI. So it's been very surprising to me that in the past year since ChatGPT has come out, we have not seen anything even sort of remotely close to that level of performance coming out of a Chinese company. Now, I do think they are working on this stuff, but it's been surprising to me that China has been mostly absent from the frontier AI conversation over the past year.

• 58:44Ezra Klein?

  And do you think those things are related? Do you think that the Chinese government's risk aversion and the underperformance of at least the products and and systems we've seen in China? I mean, there might be things we don't know about. Do you think those things are connected?

• 59:03Speaker 01

  Absolutely. I think you do need a risk appetite to be able to build and govern these systems because they are unpredictable. We don't know exactly how they work. And you know, what we saw, for example, with Microsoft was that they put out this Bing Sydney chatbot, and it got a lot of attention and and blowback, and people reported all these crazy experiences. And in China, if something like that had happened, you know, they might have shut the company down, or they might, you know, it might have been deemed such an embarrassment that they would have radically scaled back the model. And instead, what Microsoft did was just kind of say, like, we're going to make some changes to try to prevent that kind of thing from happening, but we're keeping this model out there. We're going to let the public use it, and they'll probably discover other crazy things. And

• 59:46Speaker 01

  that's just part of the learning process. That's something that I've been convinced of over the past year, talking with AI executives and people at these companies, is that you really do need some contact with the public before you start learning everything that these models are capable of and all the ways that they might misbehave.

• 1:00:05Ezra Klein?

  What is the European Union trying to do? They've had, you know, draft regulations that were seemed very expansive. What has been the difference in how they're trying to regulate this versus how we are? And and what in your understanding is the status of their effort?

• 1:00:19Speaker 02

  Europe was quite ahead with developing its AI Act, but it was written in a pre-chat GPT world. It was written in a pre-genenerative AI world. And so over the past year, they've been trying to retrofit it so that it kind of reflects our new reality and is caught up in in debate in the meantime. But my understanding is the AI Act is not particularly restrictive on on what these companies can do. So to my understanding, there's nothing in the AI Act that is going to prevent these next generation technologies from being built. It's more about companies being transparent.

• 1:00:55Speaker 01

  Let me add a little bit of flavor to that because I was in Europe just recently talking with some lawmakers and one of the things that people will say about the AI Act is that it has this risk-based framework where different AI products are sort of evaluated and regulated based on these classifications of this is a low-risk system, or this is a high-risk system, or this is a medium-risk system. And so different rules apply based on which of those buckets a new tool falls into. And so right now, what a lot of regulators and politicians and companies and lobbyists in Europe are arguing about is what level of risk should something like a foundation model, a GPT-4, a Bard, a Claude, are those

• 1:01:36Speaker 01

  low-risk systems because they're just chatbots? Or are they high-risk systems because you can build so many other things once you have that basic technology? And so that's sort of what my understanding is of the current battle in Europe is over whether foundation models, frontier models, whatever you want to call them, whether those should be assigned to one risk bucket or another.

• 1:02:01Ezra Klein?

  I think that's a good survey of the waterfront. And so I guess I'll end on this question, which is, all right, we're talking here at the one-year anniversary roughly of ChatGPT. If you were to guess, if we were having another conversation a year from now on the two-year anniversary, what do you think would have changed? What are one or two things each of you think is likely to happen over the next year that did not happen this year?

• 1:02:24Speaker 01

  I think all communication-based work will start to have an element of AI in it. All email, all presentations, office work, essentially, AI will be built into all the applications that we use for that stuff. And so it'll just be sort of part of the background, just like autocomplete is today when you're typing something on your phone.

• 1:02:44Speaker 02

  I would say that AI is going to continue to hollow out the media industry. I think you're going to see more publishers turning to these really bad services that just automate the generation of copy. You'll see more sort of content farms springing up on the web. It'll reduce publisher revenue and we'll just see more. Digital media businesses either get sold or sort of quietly go out of business. And that's going to go hand in hand with the decline of the web in general. A year from now, more and more people are going to be using chat GPT and other tools as their kind of front door to internet knowledge. And that's just going to sap a lot of life out of the web as we know it. So we don't need one more technological breakthrough for any of that to happen.

• 1:03:28Speaker 02

  That's just a case of consumer preferences taking a while to change. And I think it's well underway.

• 1:03:34Ezra Klein?

  So do you think then that next year, we're going to see something that has been long predicted, which is significant AI related job losses? Is that sort of the argument you're making here?

• 1:03:42Speaker 02

  I think that to some degree, it already happened this year in digital media. And yes, I do think it will, it will start to pick up. You know, just keep in mind, 12 months is not a lot of time for every single industry to ask itself, could I get away with five or 10 or 15% fewer employees? And as the end of this year comes around, I have to believe that in lots and lots of industries, people are going to be asking that question.

• 1:04:05Speaker 01

  Yeah, I agree. I don't know whether there will be sort of one year where all the jobs that are going to vanish, vanish. I think it's more likely to be a slow trickle over time. And, and it's less likely to be mass layoffs than just new entrants that can do the same work as incumbents with many fewer people. You know, the, the software development firm that only needs five coders because they have, you know, all their coders are using AI and they have software that is sort of building itself, competing with companies that have, you know, 10,000 engineers and, and doing so much more capably. So I don't think it's going to necessarily look like all the layoffs hit on one day or in one quarter or even in one year. But I do think we're already seeing displacement of jobs through AI.

• 1:04:48Ezra Klein?

  Those are kind of dark predictions. I mean, we'll have a little bit better sort of integration of AI into office tools. And also we'll begin to see really the, the productivity improvements create job losses. Is there anything that, that you think is coming down the pipe technologically? That would be, you know, really deeply to the good things that are not too far from fruition that you think will make life a lot better for people?

• 1:05:10Speaker 02

  I mean, I love the idea of universal translators. It's already pretty good using AI to, you know, speak in one language and get output in another. But I do think that's going to enable a lot of cross-cultural communications. And there are a lot of products remaining to be built that will essentially just drop the latency so that you can talk and hear in real time and have it be quite good. So that's something that makes me happy.

• 1:05:33Speaker 01

  And I'm hopeful that we will use AI, not we, as in me and Casey.

• 1:05:37Speaker 02

  But we might.

• 1:05:39Speaker 01

  This would be sort of a career change for us. But we, as in, as in society, I have some hope that we will use AI to cure one of the sort of top deadliest diseases, cancer, heart disease, Alzheimer's, things like that that really affect massive numbers of people. I don't have any inside reporting that we are on the cusp of a breakthrough, but I know that a lot of energy and research and funding is going into using AI to discover new drugs and therapies for some of the leading killer diseases and conditions in the world. And so when I want to feel more optimistic, I just think about the possibility that all of that bears fruit sometime in the next few years, and that's pretty exciting.

• 1:06:18Ezra Klein?

  All right. And then also a final question. What are a few books you'd each recommend to the audience, or least recommend the audience ask ChatGPT to summarize for them?

• 1:06:27Speaker 01

  Kevin, you want to go first? Sure. I actually have two books and a YouTube video. The two books, one of them is called Electrifying America by David E. Nye. It is a 30-year-old history book about the process by which America got electricity. And it has been very interesting to read. I read it first a few years ago and have been rereading it. Just to sort of sketch out, what would it look like if AI really is the new electricity? What happened the last time society was transformed by a technology like this? The other book I'll recommend is Your Face Belongs to Us by my colleague, our colleague at the Times, Kashmir Hill, which is about the facial recognition AI company ClearView AI and is one of the most compelling tech books I've read in a few years.

• 1:07:12Speaker 01

  And then the YouTube video I'll recommend was just posted a few days ago. It's called Intro to Large Language Models. It's made by Andrej Karpathy, who is an AI researcher, actually at OpenAI. And it's his one-hour introduction to what is a large language model and how does it work? And I've just found it very helpful for my own understanding.

• 1:07:32Ezra Klein?

  Casey?

• 1:07:33Speaker 02

  Well, as we're with permission, and given that Kevin has just given your listeners two great books and a YouTube video to read, I would actually like to recommend three newsletters if I could. And the reason is because the books that were published this year did not help me really understand the future of the AI industry. And to understand what's happening in real time, I really am leaning on newsletters more than I'm leaning on books. So is that okay?

• 1:07:53Speaker 01

  Yeah, go for it.

• 1:07:54Speaker 02

  All right. So the first one, cruelly, you already mentioned earlier in this podcast, it's Import AI from Jack Clark. Jack co-founded Anthropic, one of the big AI developers, and it is fascinating to know which papers he's reading every week that are helping him understand this world. And I think that they're arguably having an effect on how Anthropic is being created because he is sitting in all of those rooms. So that is just an incredible weekly read. I would also recommend AI Snake Oil from the Princeton professor Arvind Narayanan and a PhD student at Princeton, Sai Ash Kapoor. They're very skeptical of AI hype and doomsday scenarios, but they also take the technology really seriously and have a lot of smart thoughts about policy and regulation. And then the final one is Pragmatic Engineer by this guy Gergely Arroz. He's this former Uber engineering

• 1:08:36Speaker 02

  manager, and he writes about a lot of companies, but he writes about them as workplaces. And I love when he writes about OpenAI as a workplace. He interviews people there about culture and management and process, and he just constantly reminds you, they're just human beings showing up to the office every day and building this stuff. And it's just a really unique viewpoint on that world. So read those three newsletters. You'll have a little better sense of what's coming for us in the future.

• 1:08:56Speaker 01

  What Casey didn't say is that he actually hasn't read a book in 10 years. So it was a bit of a trick question.

• 1:09:01Speaker 02

  You know what I will say? I did read Your Face Belongs to Us by Cashan. Incredible book. Definitely read that one.

• 1:09:06Speaker 01

  Sure you did.

• 1:09:06Ezra Klein?

  There you go. Casey Newton, Kevin Roos, your podcast, which requires very little reading, is hard for it. Thank you all for being on the show.

• 1:09:14Speaker 01

  It's the first illiterate podcast, actually, put out by the New York Times. Thank you for having us.

• 1:09:18Speaker 02

  Thanks, Ezra.

• 1:09:19Ezra Klein?

  Thanks, guys.

• 1:09:36Ezra Klein?

  This episode of The Ezra Klein Show was produced by Roland Hoof, fact checking by Michelle Harris with Kate Sinclair and Mary Marge Locker. Our senior engineer is Jeff Geld, our senior editor is Claire Gordon. The show's production team also includes Emma Volgawu and Kristin Lin. Original music by Isaac Jones, audience strategy by Christina Samulski and Shannon Busta. The executive producer of New York Times Opinion Audio is Annie Rose Strasser. And special thanks to Sonia Herrera.